I am trying to optimize a MySQL select request:
SELECT * FROM `sales`
WHERE ((sales.private = false AND (sales.buyer_id IS NULL OR NOT sales.buyer_id=142)
AND (sales.merchand_id IS NULL OR NOT sales.merchand_id=142)
AND (sales.private_item = false) )
AND ((sales.buyer_id=32 OR sales.merchand_id=32)
AND (sales.admin=0 AND NOT sales.type IN ('book'))))
ORDER BY sales.created_at DESC, sales.id DESC LIMIT 0, 10;
The schema of the table is
mysql> SHOW columns from sales;
+------------------------+--------------+------+-----+---------+----------------+
| Field | Type | Null | Key | Default | Extra |
+------------------------+--------------+------+-----+---------+----------------+
| id | int(11) | NO | PRI | NULL | auto_increment |
| type | varchar(255) | YES | MUL | NULL | |
| buyer_id | int(11) | YES | MUL | NULL | |
| merchand_id | int(11) | YES | MUL | NULL | |
| private | tinyint(1) | YES | | 0 | |
| admin | tinyint(1) | YES | | 0 | |
| created_at | datetime | YES | | NULL | |
| updated_at | datetime | YES | | NULL | |
| country_id | int(11) | YES | MUL | 0 | |
| private_item | tinyint(1) | YES | | 0 | |
+------------------------+--------------+------+-----+---------+----------------+
The indexes are:
mysql> show indexes from sales;
+-----------------+------------+--------------------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment |
+-----------------+------------+--------------------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+
| sales | 0 | PRIMARY | 1 | id | A | 286509 | NULL | NULL | | BTREE | |
| sales | 1 | index_sales_on_type | 1 | type | A | 123 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_on_buyer_id | 1 | buyer_id | A | 40929 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_on_merchand_id | 1 | merchand_id | A | 40929 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_on_country_id | 1 | country_id | A | 6 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_on_type_and_country_id | 1 | type | A | 151 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_on_type_and_country_id | 2 | country_id | A | 428 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_viewed | 1 | buyer_id | A | 35813 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_viewed | 2 | merchand_id | A | 286509 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_viewed | 3 | private_item| A | 285009 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_viewed | 4 | admin | A | 285009 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_viewed | 5 | type | A | 285009 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_viewed | 6 | private | A | 285009 | NULL | NULL | YES | BTREE | |
| sales | 1 | index_sales_viewed | 7 | created_at | A | 285009 | NULL | NULL | YES | BTREE | |
+-------+------------+------------------------------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+
When doing the query it uses the index_sales_on_type_and_country_id even if there is no country_id in the query…
The query takes 2.5 seconds with this index.
But when I use USE INDEX(index_sales_viewed) it goes down to 0.2 seconds.
Here is the EXPLAIN of the query:
+----+-------------+-----------------+------+----------------------------------------------+------+---------+------+--------+---------------------
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-----------------+------+----------------------------------------------+------+---------+------+--------+---------------------
| 1 | SIMPLE | sales | range | see bellow | index_sales_on_type_and_country_id | 258 | NULL | 208725 | Using where; Using filesort |
+----+-------------+-----------------+------+----------------------------------------------+------+---------+------+--------+---------------------
the possible keys are :
index_sales_on_type,
index_sales_on_buyer_id,
index_sales_on_merchand_id,
index_sales_on_type_and_country_id,
index_sales_public_recent_activity
Why doesn’t MySQL use index_sales_viewed by default? Could there be a better index?
Thank you!
This is wrong use on NULL, please change all the column used in the index to be NOT NULL
refer this When to use NULL in MySQL tables
official documentation
Mysql chosen index index_sales_on_type_and_country_id because you are not compare with NULL value